TY  - RPRT
TI  - Depth Laws for the Precision Floor of Trained Neural Networks: Amplification, Residual Scaling, and a Quantization-Aware Training Paradox
AU  - Ahmad S. Tarawneh
PY  - 2026
UR  - https://arxiv.org/abs/2609.32060
ID  - 2609.32060
ER  -